Method configure

src/gpu/cl/operators/ClConcatenate.cpp:45–143

namespace opencl
{
void ClConcatenate::configure(const CLCompileContext           &compile_context,
                              const std::vector<ITensorInfo *> &src_vector,
                              ITensorInfo                      *dst,
                              size_t                            axis)
{
    ARM_COMPUTE_ERROR_ON(dst == nullptr);
    ARM_COMPUTE_LOG_PARAMS(src_vector, dst, axis);
    _axis       = axis;
    _num_inputs = src_vector.size();

    TensorShape dst_shape = arm_compute::misc::shape_calculator::calculate_concatenate_shape(src_vector, _axis);
    std::vector<const ITensorInfo *> const_src_vector(src_vector.size());
    std::transform(src_vector.begin(), src_vector.end(), const_src_vector.begin(),
                   [](ITensorInfo *t)
                   {
                       ARM_COMPUTE_ERROR_ON_NULLPTR(t);
                       return t;
                   });

    // dst auto inizialitation if not yet initialized
    auto_init_if_empty(*dst, dst_shape, 1, src_vector[0]->data_type());
    ARM_COMPUTE_ERROR_THROW_ON(ClConcatenate::validate(const_src_vector, dst, axis));

    unsigned int offset = 0;
    switch (_axis)
    {
        case Window::DimX:
        {
            switch (_num_inputs)
            {
                case 2:
                {
                    // Configure WidthConcatenate2Tensors kernel
                    auto kernel = std::make_unique<kernels::ClWidthConcatenate2TensorsKernel>();
                    kernel->configure(compile_context, src_vector.at(0), src_vector.at(1), dst);
                    _concat_kernels.emplace_back(std::move(kernel));
                    break;
                }
                case 4:
                {
                    // Configure WidthConcatenate4Tensors kernel
                    auto kernel = std::make_unique<kernels::ClWidthConcatenate4TensorsKernel>();
                    kernel->configure(compile_context, src_vector.at(0), src_vector.at(1), src_vector.at(2),
                                      src_vector.at(3), dst);
                    _concat_kernels.emplace_back(std::move(kernel));
                    break;
                }
                default:
                {
                    // Configure generic case WidthConcatenate kernels
                    for (unsigned int i = 0; i < _num_inputs; ++i)
                    {
                        auto kernel = std::make_unique<kernels::ClWidthConcatenateKernel>();
                        kernel->configure(compile_context, src_vector.at(i), offset, dst);
                        offset += src_vector.at(i)->dimension(_axis);
                        _concat_kernels.emplace_back(std::move(kernel));
                    }
                    break;